Classification of Traditional and Modern Music using NCC and k-NN

Abstract

: Music is a means of interaction between humans which is transmitted as a presentation of feelings through acoustic sensation. Music consists of instruments played ensemble, occasionally with vocal, to form a harmony. The presence of certain instruments can be used to identify the genre of a music, and in turn its origin. This research conducted classification of traditional, local contemporary, and foreign music – from Indonesian point of view – according to instruments and beats. Genres chosen to represent the music in this research, fall into six categories: Balinese, Javanese, Sundanese (traditional), Keroncong (local contemporary), Classical and Latin (foreign). 180 pieces of music are used for training, and the same number of pieces are used for testing; using samples of pieces with instruments only and also instruments with vocal. To extract its features, each music pieces are cut into 30ms slices, then a representative vector of 3 time-domain features is taken from every piece. Classification of test data is then conducted using Nearest Centroid Classifier (NCC) and k-Nearest Neighbour (k-NN) with k=3 and k=5. Best results are obtained using k-NN with k=3, generating the maximum 96.6% accuracy for Balinese, with an all-genre average of 73.89%. The lowest accuracy rate belongs to Classical category, in which from the three tests, it is consistently rated under 50% with average of 36.67%.